Growing recognition of the social determinants of health has stimulated research on social capital and mental health. In recent years, systematic reviews using cross-sectional data have found that social capital may be a useful factor in the prevention of mental illness [1], [2]. In addition, a prospective study by Fujiwara and Kawachi [3] showed that a low level of social capital, measured by trust, is associated with depression.

As for measuring social capital, these previous studies have provided evidence for the importance of distinguishing the structural components of social capital (structural social capital) from its cognitive components (cognitive social capital). Structural social capital refers to what people do (e.g., participation in associations) while cognitive social capital refers to what people feel (e.g., trust in others, reciprocity between individuals) [4]. These two components may differentially affect mental health outcomes. For instance, a systematic review by De Silva [2] found that although cognitive social capital is associated with better mental health, structural social capital is associated with poor mental health.

Despite these previous studies, evidence on the association between social capital and mental health is limited by the fact that there has been little empirical research that adopts a multilevel framework to assess whether social capital at the ecological level is associated with individual mental health. That is, a multilevel model implies that variations in mental health outcomes are determined by compositional effects (e.g., age, sex, educational attainment, income) as well as by contextual effects such as community social capital [5]. Social capital at the ecological level has most often been measured by aggregating individual perceptions of social capital. However, the effect of social capital at the ecological level on mental health might be confounded by individual perceptions of social capital. We therefore need to understand whether social capital at the ecological level exerts a contextual effect on mental health, even after adjusting for individual perceptions of social capital.

In the present study, we defined community social capital based on the “cho-cho” or “aza” unit, which is made up of approximately 250 households. The aim of this study was to examine (1) whether cognitive and structural social capital at the ecological level were associated with individual mental health by means of a multilevel analysis; (2) whether social capital at the ecological level exert a contextual effect on mental health after individual perceptions of social capital were taken into account.

Table 1 provides a summary of the data for the analysis. The mean score of mental health was 65.1 (standard deviation: 18.3), ranging from 0.0 to 100.0. Regarding demographic characteristics, close to half of the sample were male (50.7%), and 29.3% were 60–69 years of age, followed by 50–59 years of age (18.8%), and 70–79 years of age (17.7%). As for socioeconomic characteristics, 42.9% were high school graduates, and about 30.0% had more than university-level of education. 19.6% had household incomes of over 8 million yen and 14.0% had incomes under 2 million yen. In terms of social capital at neighborhood level, the mean percentage of reporting high trust was 56.6% (standard deviation: 12.7). The mean percentage of respondents active in neighborhood associations was 49.7% (standard deviation: 17.0); in sports, hobby, recreation, or cultural groups was 36.0% (standard deviation: 10.1).

Table 2 provides the results of the multilevel analyses. The null model with no predictors (Model 1) revealed a significant variation in mental health between neighborhoods (σ²_u0 = 4.669). However, this result did not take into account the compositional characteristics. In Model 2, females were more likely to have lower mental health scores. For the age variables, respondents 50–59 years of age or older were more likely to have higher mental health scores. As for the socioeconomic variables, those having an income of 2.0 million yen or more were likely to have higher mental health scores. Further, those with educational attainments equal to or higher than high school education were more likely to have higher mental health scores.

In Model 3, we observed that higher scores of cognitive social capital, measured by trust, as well as structural social capital, measured by membership in sports, recreation, hobby, or cultural groups, were more likely to have higher mental health scores, after adjustment for individual confounders (Table 2, Models 3A and 3C). Finally, after adjusting for social capital perceptions at the individual level, we found that the association between neighborhood social capital and mental health also remained (Table 3, Models 4A and 4C).

This study suggests that cognitive social capital as well as structural social capital, measured by membership in sports, recreation, hobby, or cultural groups, were associated with individual mental health status (Models 3A and 3C). Specifically, we focused on social capital at the ecological level in a geographically defined neighborhood, and our findings advance the empirical evidence on social capital and mental health within small area units. In addition, our findings enabled us to answer the question: do neighborhood levels of social capital exert a contextual effect on mental health after adjustment for perceptions of social capital? We found that both cognitive social capital and structural social capital appeared to have a contextual influence on mental health in Japan (Models 4A and 4C). Furthermore, since we developed the statistical models sequentially, we could examine changes in the amount of neighborhood variance in mental health explained by the variables in the regression. For example, comparing Models 1 and 3A, the neighborhood variances were reduced. This result suggested that the variation found by Model 1 was explained by differences in the compositional factors and contextual factors.

Our findings are consistent with some previous individual-level and ecological-level studies on social capital and mental health. For example, the literature review by De Silva et al. [2] reported that cognitive social capital is associated with better mental health. In addition, a multilevel study in a rural population found that cognitive social capital, measured by trust was, positively associated with psychological health [21]. On the other hand, Stafford et al. [22] reported that there was no evidence of an association between cognitive social capital, measured by trust, and mental health based on multilevel analyses in the United Kingdom. Thus, these empirical findings implied that further comparative studies were needed to investigate the main effects of social capital, measured by trust, on mental health for the sake of drawing robust conclusions.

Other components of social capital, measured by membership in sports, hobby, recreation, or cultural groups, showed a statistically significant association with mental health (Model 3C). On the other hand, structural social capital, measured by membership in neighborhood associations, was not significant (Model 3B). The results of the systematic reviews showed that structural social capital, although associated with better mental health, was sometimes found to be associated with poorer mental health [2]. A possible explanation for this different statistical result between Models 3B and 3C could be that people who belong to neighborhood associations sometimes may feel burdened or suffer from a sense of duty, such as the need to attend meetings regularly or to assume managerial or supervisory roles for neighborhood-based projects. Indeed, in order to lessen the burden, neighborhood associations in recent years have often selected members from different households based on a rotating system to serve in a supervisory capacity. Thus membership in such associations is less voluntary and members might experience fewer benefits than would those who voluntarily decide to become members of sports, hobby, recreation, or cultural groups. A previous study by Kondo et al. [23] also reported that these kinds of non-cohesive community activities may be harmful to the health of the members. This hypothesis, of course, requires further in-depth investigation.

Our study had several limitations. First, the present study used a cross-sectional design, so that we could not establish the temporal order of causality. In other words, the association between civic association membership and better mental health might have reflect reverse causation, i.e. the fact that individuals with better mental health status participated in groups, rather than the other way round (participation leading to improved mental health). Second, both our outcome variable and social capital variable were self-reported. If depressed individuals are less likely to rate the trustworthiness of their neighbors in a positive light, this would lead to common method bias. Third, although our questions on social capital were developed on the basis of previous studies, the validity of the individual items (such as inquiring about perceptions of trust) is not established, and further research is required to improve the measurement of community social capital. In addition, the mean score of mental health was lower than that of a previous representative study in Japan (score was 71.6) [18]. There is still a possibility that the raw response rate might cause a potential bias in measuring mental health. Caution therefore is warranted in over-interpreting these findings. Fourth, our overall response rate was low (although not atypical postal surveys of this type). If poor mental health subjects with low social capital tended to refuse to participate in our survey, this selection bias may have had some effect on the resulting association between social capital and mental health. In addition, the majority of participants were older and highly educated. Thus, further research should compare results from different generations (e.g. old and young populations) or socioeconomic status (high and low educational attainment). Fifth, there may be a difference between urban and rural populations [21]. While developing this challenge was beyond the scope of our paper, we consider this as a potentially important issue for further exploration.

Our study also has some strengths. To our knowledge, it is the first study of social capital and mental health in Japan that utilized a nationally representative sample. Our areal unit for measuring neighborhood social capital was based on prior theory that incorporates relevant sociodemographic characteristics of residents. We measured and tested two separate dimensions of social capital – the cognitive and the structural – and we tested them in a multilevel analytical framework to rule out the influence of confounding by compositional factors. Our findings suggest that the concept of community social capital may have relevance for mental health promotion in Japanese society.